The proposal features a new approach to address two questions of considerable policy interest: How do physician reimbursement rules affect clinical decision making? And how do key clinical decisions affect actual patient outcomes? The key insight is that using econometric methods to address both questions simultaneously in effect creates very large "natural" clinical trials to test the effectiveness of many procedures that are commonly used in the Medicare population. First, the project will use detailed, comprehensive datasets on Medicare beneficiary utilization to estimate the effects of changes in physician reimbursement for the so called "overpriced" procedures on clinical management decisions. This analysis required the initial development of computational and data management techniques for using all components of the Medicare beneficiary linked files for 1986-90 as well as a dataset including all Medicare patients who had new myocardial infarctions in 1987-89. Using these datasets, clinical decision trees will be developed for patients with conditions placing them at increased risk for undergoing either an "overpriced" procedure or alternative diagnostic and therapeutic management. This process will involve a combination of clinical literature review, empirical analysis of patterns of patient management, and consultations with clinical experts. The resulting set of physician choice models will be used to estimate the responsiveness of clinical decision making to exogenous variation in fees and other supply and demand factors. These supply and demand variations influence the likelihood of procedure choices for individual patients for reasons independent of the patient's underlying disease severity and other characteristics that would be correlated with both procedure choice and later clinical outcomes. Consequently, the variations are instrumental variables that randomize patients to alternative treatments for nonclinical reasons. With econometric techniques that make use of this fact, the project will estimate the impact of treatment choice on patient outcomes without bias due to underlying population differences. Though applied to only a limited set of procedures and clinical problems here, these methods are potentially very widely applicable to outcomes research and to studies of the effects of changes in physician incentives such as the recent fundamental reforms in the Medicare physician fee schedule.